nengel@2: /*
nengel@2:  * copyright (c) 2004 AGAWA Koji
nengel@2:  *
nengel@2:  * This file is part of FFmpeg.
nengel@2:  *
nengel@2:  * FFmpeg is free software; you can redistribute it and/or
nengel@2:  * modify it under the terms of the GNU Lesser General Public
nengel@2:  * License as published by the Free Software Foundation; either
nengel@2:  * version 2.1 of the License, or (at your option) any later version.
nengel@2:  *
nengel@2:  * FFmpeg is distributed in the hope that it will be useful,
nengel@2:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
nengel@2:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
nengel@2:  * Lesser General Public License for more details.
nengel@2:  *
nengel@2:  * You should have received a copy of the GNU Lesser General Public
nengel@2:  * License along with FFmpeg; if not, write to the Free Software
nengel@2:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
nengel@2:  */
nengel@2: 
nengel@2: #include "libavcodec/avcodec.h"
nengel@2: #include "libavcodec/dsputil.h"
nengel@2: #include "libavcodec/mpegvideo.h"
nengel@2: #include "mpegvideo_arm.h"
nengel@2: 
nengel@2: static void dct_unquantize_h263_intra_iwmmxt(MpegEncContext *s,
nengel@2:                                              DCTELEM *block, int n, int qscale)
nengel@2: {
nengel@2:     int level, qmul, qadd;
nengel@2:     int nCoeffs;
nengel@2:     DCTELEM *block_orig = block;
nengel@2: 
nengel@2:     assert(s->block_last_index[n]>=0);
nengel@2: 
nengel@2:     qmul = qscale << 1;
nengel@2: 
nengel@2:     if (!s->h263_aic) {
nengel@2:         if (n < 4)
nengel@2:             level = block[0] * s->y_dc_scale;
nengel@2:         else
nengel@2:             level = block[0] * s->c_dc_scale;
nengel@2:         qadd = (qscale - 1) | 1;
nengel@2:     }else{
nengel@2:         qadd = 0;
nengel@2:         level = block[0];
nengel@2:     }
nengel@2:     if(s->ac_pred)
nengel@2:         nCoeffs=63;
nengel@2:     else
nengel@2:         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
nengel@2: 
nengel@2:     __asm__ volatile (
nengel@2: /*      "movd %1, %%mm6                 \n\t" //qmul */
nengel@2: /*      "packssdw %%mm6, %%mm6          \n\t" */
nengel@2: /*      "packssdw %%mm6, %%mm6          \n\t" */
nengel@2:         "tbcsth wr6, %[qmul]            \n\t"
nengel@2: /*      "movd %2, %%mm5                 \n\t" //qadd */
nengel@2: /*      "packssdw %%mm5, %%mm5          \n\t" */
nengel@2: /*      "packssdw %%mm5, %%mm5          \n\t" */
nengel@2:         "tbcsth wr5, %[qadd]            \n\t"
nengel@2:         "wzero wr7                      \n\t" /* "pxor %%mm7, %%mm7             \n\t" */
nengel@2:         "wzero wr4                      \n\t" /* "pxor %%mm4, %%mm4             \n\t" */
nengel@2:         "wsubh wr7, wr5, wr7            \n\t" /* "psubw %%mm5, %%mm7            \n\t" */
nengel@2:         "1:                             \n\t"
nengel@2:         "wldrd wr2, [%[block]]          \n\t" /* "movq (%0, %3), %%mm0          \n\t" */
nengel@2:         "wldrd wr3, [%[block], #8]      \n\t" /* "movq 8(%0, %3), %%mm1         \n\t" */
nengel@2:         "wmulsl wr0, wr6, wr2           \n\t" /* "pmullw %%mm6, %%mm0           \n\t" */
nengel@2:         "wmulsl wr1, wr6, wr3           \n\t" /* "pmullw %%mm6, %%mm1           \n\t" */
nengel@2: /*      "movq (%0, %3), %%mm2           \n\t" */
nengel@2: /*      "movq 8(%0, %3), %%mm3          \n\t" */
nengel@2:         "wcmpgtsh wr2, wr4, wr2         \n\t" /* "pcmpgtw %%mm4, %%mm2          \n\t" // block[i] < 0 ? -1 : 0 */
nengel@2:         "wcmpgtsh wr3, wr4, wr2         \n\t" /* "pcmpgtw %%mm4, %%mm3          \n\t" // block[i] < 0 ? -1 : 0 */
nengel@2:         "wxor wr0, wr2, wr0             \n\t" /* "pxor %%mm2, %%mm0             \n\t" */
nengel@2:         "wxor wr1, wr3, wr1             \n\t" /* "pxor %%mm3, %%mm1             \n\t" */
nengel@2:         "waddh wr0, wr7, wr0            \n\t" /* "paddw %%mm7, %%mm0            \n\t" */
nengel@2:         "waddh wr1, wr7, wr1            \n\t" /* "paddw %%mm7, %%mm1            \n\t" */
nengel@2:         "wxor wr2, wr0, wr2             \n\t" /* "pxor %%mm0, %%mm2             \n\t" */
nengel@2:         "wxor wr3, wr1, wr3             \n\t" /* "pxor %%mm1, %%mm3             \n\t" */
nengel@2:         "wcmpeqh wr0, wr7, wr0          \n\t" /* "pcmpeqw %%mm7, %%mm0          \n\t" // block[i] == 0 ? -1 : 0 */
nengel@2:         "wcmpeqh wr1, wr7, wr1          \n\t" /* "pcmpeqw %%mm7, %%mm1          \n\t" // block[i] == 0 ? -1 : 0 */
nengel@2:         "wandn wr0, wr2, wr0            \n\t" /* "pandn %%mm2, %%mm0            \n\t" */
nengel@2:         "wandn wr1, wr3, wr1            \n\t" /* "pandn %%mm3, %%mm1            \n\t" */
nengel@2:         "wstrd wr0, [%[block]]          \n\t" /* "movq %%mm0, (%0, %3)          \n\t" */
nengel@2:         "wstrd wr1, [%[block], #8]      \n\t" /* "movq %%mm1, 8(%0, %3)         \n\t" */
nengel@2:         "add %[block], %[block], #16    \n\t" /* "addl $16, %3                  \n\t" */
nengel@2:         "subs %[i], %[i], #1            \n\t"
nengel@2:         "bne 1b                         \n\t" /* "jng 1b                                \n\t" */
nengel@2:         :[block]"+r"(block)
nengel@2:         :[i]"r"((nCoeffs + 8) / 8), [qmul]"r"(qmul), [qadd]"r"(qadd)
nengel@2:         :"memory");
nengel@2: 
nengel@2:     block_orig[0] = level;
nengel@2: }
nengel@2: 
nengel@2: #if 0
nengel@2: static void dct_unquantize_h263_inter_iwmmxt(MpegEncContext *s,
nengel@2:                                              DCTELEM *block, int n, int qscale)
nengel@2: {
nengel@2:     int nCoeffs;
nengel@2: 
nengel@2:     assert(s->block_last_index[n]>=0);
nengel@2: 
nengel@2:     if(s->ac_pred)
nengel@2:         nCoeffs=63;
nengel@2:     else
nengel@2:         nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
nengel@2: 
nengel@2:     ippiQuantInvInter_Compact_H263_16s_I(block, nCoeffs+1, qscale);
nengel@2: }
nengel@2: #endif
nengel@2: 
nengel@2: void MPV_common_init_iwmmxt(MpegEncContext *s)
nengel@2: {
nengel@2:     if (!(mm_flags & FF_MM_IWMMXT)) return;
nengel@2: 
nengel@2:     s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_iwmmxt;
nengel@2: #if 0
nengel@2:     s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_iwmmxt;
nengel@2: #endif
nengel@2: }